DataTone: Managing Ambiguity in Natural Language Interfaces for Data Visualization
summary
DataTone is a mixed‑initiative NL→visualization system that parses queries with the Stanford parser, generates hundreds of SQL-backed interpretations, ranks them with heuristics, and surfaces choices with ambiguity widgets so users set constraints that persist across queries. The prototype produces 576 data specifications and over 5,000 visualizations for a simple query, runs on a single-table Olympic dataset, and in a 16-participant Jeopardy-style study 14 participants prefer it to IBM Watson Analytics while completing more facts correctly.
AI-written summary · about
chapters (7)
- 0:00Introduction and motivation
- 1:07Limitations of existing NLI tools
- 3:35DataTone prototype demonstration
- 7:25How DataTone resolves ambiguity
- 9:54Evaluation methodology and results
- 13:22Limitations and future work
- 14:45Conclusion and Q&A
related talks
20mEviza: A Natural Language Interface for Visual Analysis
16mType-Directed Synthesis of Visualizations from Natural Language Queries
42mInteractive Data Analysis: Visualization and Beyond
02mDBTB INT Aaron Kalb r
01h 00mData, visualization, and designing AI
04mDBTB INT Akash Mukherjee
32mData Viz for Engineers: Optimizing Insight & Decision Making Through Visualization
18mConsidering Agency and Data Granularity in the Design of Visualization Tools